Build and Compare an ENA Model
Configure units, conversations, codes, and a stanza window in webENA, then compare networks without confusing visual separation with statistical evidence.
Der geprüfte Tutorialtext ist derzeit auf Englisch verfügbar.
Methodenquellen
Vollständiges Tutorial
A reproducible ENA model begins with explicit choices. The software needs to know which rows form each network, where relational context begins and ends, which columns are codes, and how nearby rows may contribute to a connection. Those settings operationalize the research design from Lesson 1 and the table structure from Lesson 2.
This tutorial uses the official webENA workflow as the main path. Interface labels may evolve, but the analytic record should remain stable: dataset version, units, conversation fields, codes, window, weighting, normalization, rotation, groups, and exported outputs.
Lehrbeispiel
Eight fictional teacher-design teams
The teaching CSV contains eight team units, four baseline and four scaffolded. Each team has one ordered design-review conversation and five binary code columns: Goal, Evidence, Strategy, Tradeoff, and Revision. The values were authored to make the workflow visible, not to simulate a population.
Übungsdatensatz
Schritt 1
Create a set and inspect the import
Open the official webENA application, create a project or set, and upload the teaching CSV. Confirm that the header row is recognized, rows remain in line-number order, and the raw utterance column is available for later evidence inspection. Do not proceed if unit, conversation, or code columns have missing values.
PrüfpunktThe imported table has 48 ordered rows and the expected metadata and five binary code columns.Schritt 2
Define the units
Select team_id as the unit field so each team becomes one network. Keep condition as metadata for grouping rather than including it in the unique unit identifier for this teaching model. In a real study, use every field required to uniquely identify the analytic unit across the dataset.
PrüfpunktThe model reports eight units, with four baseline and four scaffolded teams.Schritt 3
Bound the conversation
Use conversation_id as the conversation field and retain discussion_round as readable metadata plus line_number as the within-conversation order. The supplied conversation_id combines team_id and discussion_round so it is unique across the dataset. If your ENA interface accepts multiple conversation fields, selecting team_id and discussion_round together is equivalent. Choose a moving stanza window that includes the current line and up to four previous lines for the first run. This is a pedagogical choice for six-line conversations, not a universal default. Record the window and justify it from the interaction process.
PrüfpunktEach conversation_id belongs to exactly one team, connections cannot cross team or discussion-round boundaries, and the recorded window matches the model setting.Schritt 4
Select the code columns
Select goal, evidence, strategy, tradeoff, and revision as codes. Keep identifiers, condition, speaker, line number, and utterance as metadata. Confirm that the model-validity checklist is complete before reading a blank plot as a result.
PrüfpunktExactly five intended code nodes appear and no metadata column is plotted as a code.Schritt 5
Build the shared analytic space
Generate the model and keep all eight units in the same space. Inspect the comparison plot, individual points, network graphs, axis variance, and line-weight display. Save a screenshot or export together with the complete settings record. Node placement helps make connection patterns visible; it is not a geographic map of the concepts.
PrüfpunktOne documented model contains all units and retains coordinated point and network views.Schritt 6
Compare conditions cautiously
Group units by condition and add the baseline and scaffolded means to the same comparison plot. Inspect their mean networks and the comparison network. If you run statistical tools, label the analysis as a demonstration with a tiny synthetic sample. A visible distance or thick subtraction edge is a prompt for inspection, not proof of a reliable population difference.
PrüfpunktYour notes distinguish the visual comparison from any inferential test and retain the eight-unit sample size.